aoTuV b6.03 oggenc+libvorbis slightly diff outputs; why/which is best?, [TOS #6: was "Ogg/Vorbis Lancer builds"] Options

[Anakunda]

HI!!
As you know there's published new oggenc and libvorbis Lancer builds based on aoTuV b6.03. So regarding it's writing library they should use the same code, however looking closer on the result the Lancer builds give slightly different output:

Comparing:
"C:\Users\Anakunda\AppData\Local\Temp\aoTuVb6.03.ogg"
"C:\Users\Anakunda\AppData\Local\Temp\Lancer.ogg"
Differences found: 1179627 sample(s), starting at 0.4542404 second(s), peak: 0.1180483 at 30.8317687 second(s), 2ch

So obviously I'm interested where's the difference and which builds are expected to give more transparent output. Thanks!

[john33]

The speed up in the 'Lancer' builds is directly as a consequence of replacing the standard code with highly optimized 'hand-rolled' SSE/SSE2 specific code. It is not simply the use of a different set of compiler options. Two particular routines that are heavily modified are 'psy.c' (increases from 2211 lines of code to 5178) and 'smallft.c' (increases from 1256 lines of code to 6832). Output differences are, therefore, to be expected although no one has suggested they are audible (except in a particular instance referred to elsewhere).

I don't recall hearing tests having been conducted, but I also don't recall anyone claiming to hear any differences (save for the instance already mentioned).